Merge branch 'nd/untracked-cache'
authorJunio C Hamano <gitster@pobox.com>
Tue, 26 May 2015 20:24:45 +0000 (13:24 -0700)
committerJunio C Hamano <gitster@pobox.com>
Tue, 26 May 2015 20:24:46 +0000 (13:24 -0700)
Teach the index to optionally remember already seen untracked files
to speed up "git status" in a working tree with tons of cruft.

* nd/untracked-cache: (24 commits)
git-status.txt: advertisement for untracked cache
untracked cache: guard and disable on system changes
mingw32: add uname()
t7063: tests for untracked cache
update-index: test the system before enabling untracked cache
update-index: manually enable or disable untracked cache
status: enable untracked cache
untracked-cache: temporarily disable with $GIT_DISABLE_UNTRACKED_CACHE
untracked cache: mark index dirty if untracked cache is updated
untracked cache: print stats with $GIT_TRACE_UNTRACKED_STATS
untracked cache: avoid racy timestamps
read-cache.c: split racy stat test to a separate function
untracked cache: invalidate at index addition or removal
untracked cache: load from UNTR index extension
untracked cache: save to an index extension
ewah: add convenient wrapper ewah_serialize_strbuf()
untracked cache: don't open non-existent .gitignore
untracked cache: mark what dirs should be recursed/saved
untracked cache: record/validate dir mtime and reuse cached output
untracked cache: make a wrapper around {open,read,close}dir()
...

21 files changed:
.gitignore
Documentation/git-status.txt
Documentation/git-update-index.txt
Documentation/technical/index-format.txt
Makefile
builtin/commit.c
builtin/update-index.c
cache.h
compat/mingw.c
compat/mingw.h
dir.c
dir.h
ewah/ewah_io.c
ewah/ewok.h
git-compat-util.h
read-cache.c
split-index.c
t/t7063-status-untracked-cache.sh [new file with mode: 0755]
test-dump-untracked-cache.c [new file with mode: 0644]
unpack-trees.c
wt-status.c
index a05241916c9c9a3760a6e98670a7f6427d553d77..422c5382c1acfde24223fda6236dd3586456a263 100644 (file)
 /test-delta
 /test-dump-cache-tree
 /test-dump-split-index
+/test-dump-untracked-cache
 /test-scrap-cache-tree
 /test-genrandom
 /test-hashmap
index 5221f950ce06cda1a764424ec217760d044a5ac3..335f3123353482cfd708420dac3b766f181e89ff 100644 (file)
@@ -66,7 +66,10 @@ When `-u` option is not used, untracked files and directories are
 shown (i.e. the same as specifying `normal`), to help you avoid
 forgetting to add newly created files.  Because it takes extra work
 to find untracked files in the filesystem, this mode may take some
-time in a large working tree.  You can use `no` to have `git status`
+time in a large working tree.
+Consider enabling untracked cache and split index if supported (see
+`git update-index --untracked-cache` and `git update-index
+--split-index`), Otherwise you can use `no` to have `git status`
 return more quickly without showing untracked files.
 +
 The default can be changed using the status.showUntrackedFiles
index aff01798cdf6b114009eae8dfc4f2866a8a24d17..1a296bc29a16fcf4ee6b581310033861c1ec82bc 100644 (file)
@@ -170,6 +170,20 @@ may not support it yet.
        the shared index file. This mode is designed for very large
        indexes that take a significant amount of time to read or write.
 
+--untracked-cache::
+--no-untracked-cache::
+       Enable or disable untracked cache extension. This could speed
+       up for commands that involve determining untracked files such
+       as `git status`. The underlying operating system and file
+       system must change `st_mtime` field of a directory if files
+       are added or deleted in that directory.
+
+--force-untracked-cache::
+       For safety, `--untracked-cache` performs tests on the working
+       directory to make sure untracked cache can be used. These
+       tests can take a few seconds. `--force-untracked-cache` can be
+       used to skip the tests.
+
 \--::
        Do not interpret any more arguments as options.
 
index 35112e4966f9b021b0889d734d5cb945b8943003..b7093af8b23e6a83741b81678e7aae0c5ff7c88a 100644 (file)
@@ -233,3 +233,65 @@ Git index format
   The remaining index entries after replaced ones will be added to the
   final index. These added entries are also sorted by entry name then
   stage.
+
+== Untracked cache
+
+  Untracked cache saves the untracked file list and necessary data to
+  verify the cache. The signature for this extension is { 'U', 'N',
+  'T', 'R' }.
+
+  The extension starts with
+
+  - A sequence of NUL-terminated strings, preceded by the size of the
+    sequence in variable width encoding. Each string describes the
+    environment where the cache can be used.
+
+  - Stat data of $GIT_DIR/info/exclude. See "Index entry" section from
+    ctime field until "file size".
+
+  - Stat data of core.excludesfile
+
+  - 32-bit dir_flags (see struct dir_struct)
+
+  - 160-bit SHA-1 of $GIT_DIR/info/exclude. Null SHA-1 means the file
+    does not exist.
+
+  - 160-bit SHA-1 of core.excludesfile. Null SHA-1 means the file does
+    not exist.
+
+  - NUL-terminated string of per-dir exclude file name. This usually
+    is ".gitignore".
+
+  - The number of following directory blocks, variable width
+    encoding. If this number is zero, the extension ends here with a
+    following NUL.
+
+  - A number of directory blocks in depth-first-search order, each
+    consists of
+
+    - The number of untracked entries, variable width encoding.
+
+    - The number of sub-directory blocks, variable width encoding.
+
+    - The directory name terminated by NUL.
+
+    - A number of untrached file/dir names terminated by NUL.
+
+The remaining data of each directory block is grouped by type:
+
+  - An ewah bitmap, the n-th bit marks whether the n-th directory has
+    valid untracked cache entries.
+
+  - An ewah bitmap, the n-th bit records "check-only" bit of
+    read_directory_recursive() for the n-th directory.
+
+  - An ewah bitmap, the n-th bit indicates whether SHA-1 and stat data
+    is valid for the n-th directory and exists in the next data.
+
+  - An array of stat data. The n-th data corresponds with the n-th
+    "one" bit in the previous ewah bitmap.
+
+  - An array of SHA-1. The n-th SHA-1 corresponds with the n-th "one" bit
+    in the previous ewah bitmap.
+
+  - One NUL.
index 25a453bf2bfd1e17c3fee1ce6697f268804401fc..323c401e966eb8a69057e563fc43b4f87e7d897f 100644 (file)
--- a/Makefile
+++ b/Makefile
@@ -574,6 +574,7 @@ TEST_PROGRAMS_NEED_X += test-date
 TEST_PROGRAMS_NEED_X += test-delta
 TEST_PROGRAMS_NEED_X += test-dump-cache-tree
 TEST_PROGRAMS_NEED_X += test-dump-split-index
+TEST_PROGRAMS_NEED_X += test-dump-untracked-cache
 TEST_PROGRAMS_NEED_X += test-genrandom
 TEST_PROGRAMS_NEED_X += test-hashmap
 TEST_PROGRAMS_NEED_X += test-index-version
index d6515a2a50e78e9376f1b55fa6fe79b1e867f25c..254477fd1d4e8b96f50eb42dc11ec1253c7467f8 100644 (file)
@@ -1366,13 +1366,14 @@ int cmd_status(int argc, const char **argv, const char *prefix)
        refresh_index(&the_index, REFRESH_QUIET|REFRESH_UNMERGED, &s.pathspec, NULL, NULL);
 
        fd = hold_locked_index(&index_lock, 0);
-       if (0 <= fd)
-               update_index_if_able(&the_index, &index_lock);
 
        s.is_initial = get_sha1(s.reference, sha1) ? 1 : 0;
        s.ignore_submodule_arg = ignore_submodule_arg;
        wt_status_collect(&s);
 
+       if (0 <= fd)
+               update_index_if_able(&the_index, &index_lock);
+
        if (s.relative_paths)
                s.prefix = prefix;
 
index 0665b31ea1e59734a2ace701b613fffebd074dde..7431938fa654ba9af524a7018ab2c7be8242caaa 100644 (file)
@@ -33,6 +33,7 @@ static int mark_valid_only;
 static int mark_skip_worktree_only;
 #define MARK_FLAG 1
 #define UNMARK_FLAG 2
+static struct strbuf mtime_dir = STRBUF_INIT;
 
 __attribute__((format (printf, 1, 2)))
 static void report(const char *fmt, ...)
@@ -48,6 +49,166 @@ static void report(const char *fmt, ...)
        va_end(vp);
 }
 
+static void remove_test_directory(void)
+{
+       if (mtime_dir.len)
+               remove_dir_recursively(&mtime_dir, 0);
+}
+
+static const char *get_mtime_path(const char *path)
+{
+       static struct strbuf sb = STRBUF_INIT;
+       strbuf_reset(&sb);
+       strbuf_addf(&sb, "%s/%s", mtime_dir.buf, path);
+       return sb.buf;
+}
+
+static void xmkdir(const char *path)
+{
+       path = get_mtime_path(path);
+       if (mkdir(path, 0700))
+               die_errno(_("failed to create directory %s"), path);
+}
+
+static int xstat_mtime_dir(struct stat *st)
+{
+       if (stat(mtime_dir.buf, st))
+               die_errno(_("failed to stat %s"), mtime_dir.buf);
+       return 0;
+}
+
+static int create_file(const char *path)
+{
+       int fd;
+       path = get_mtime_path(path);
+       fd = open(path, O_CREAT | O_RDWR, 0644);
+       if (fd < 0)
+               die_errno(_("failed to create file %s"), path);
+       return fd;
+}
+
+static void xunlink(const char *path)
+{
+       path = get_mtime_path(path);
+       if (unlink(path))
+               die_errno(_("failed to delete file %s"), path);
+}
+
+static void xrmdir(const char *path)
+{
+       path = get_mtime_path(path);
+       if (rmdir(path))
+               die_errno(_("failed to delete directory %s"), path);
+}
+
+static void avoid_racy(void)
+{
+       /*
+        * not use if we could usleep(10) if USE_NSEC is defined. The
+        * field nsec could be there, but the OS could choose to
+        * ignore it?
+        */
+       sleep(1);
+}
+
+static int test_if_untracked_cache_is_supported(void)
+{
+       struct stat st;
+       struct stat_data base;
+       int fd, ret = 0;
+
+       strbuf_addstr(&mtime_dir, "mtime-test-XXXXXX");
+       if (!mkdtemp(mtime_dir.buf))
+               die_errno("Could not make temporary directory");
+
+       fprintf(stderr, _("Testing "));
+       atexit(remove_test_directory);
+       xstat_mtime_dir(&st);
+       fill_stat_data(&base, &st);
+       fputc('.', stderr);
+
+       avoid_racy();
+       fd = create_file("newfile");
+       xstat_mtime_dir(&st);
+       if (!match_stat_data(&base, &st)) {
+               close(fd);
+               fputc('\n', stderr);
+               fprintf_ln(stderr,_("directory stat info does not "
+                                   "change after adding a new file"));
+               goto done;
+       }
+       fill_stat_data(&base, &st);
+       fputc('.', stderr);
+
+       avoid_racy();
+       xmkdir("new-dir");
+       xstat_mtime_dir(&st);
+       if (!match_stat_data(&base, &st)) {
+               close(fd);
+               fputc('\n', stderr);
+               fprintf_ln(stderr, _("directory stat info does not change "
+                                    "after adding a new directory"));
+               goto done;
+       }
+       fill_stat_data(&base, &st);
+       fputc('.', stderr);
+
+       avoid_racy();
+       write_or_die(fd, "data", 4);
+       close(fd);
+       xstat_mtime_dir(&st);
+       if (match_stat_data(&base, &st)) {
+               fputc('\n', stderr);
+               fprintf_ln(stderr, _("directory stat info changes "
+                                    "after updating a file"));
+               goto done;
+       }
+       fputc('.', stderr);
+
+       avoid_racy();
+       close(create_file("new-dir/new"));
+       xstat_mtime_dir(&st);
+       if (match_stat_data(&base, &st)) {
+               fputc('\n', stderr);
+               fprintf_ln(stderr, _("directory stat info changes after "
+                                    "adding a file inside subdirectory"));
+               goto done;
+       }
+       fputc('.', stderr);
+
+       avoid_racy();
+       xunlink("newfile");
+       xstat_mtime_dir(&st);
+       if (!match_stat_data(&base, &st)) {
+               fputc('\n', stderr);
+               fprintf_ln(stderr, _("directory stat info does not "
+                                    "change after deleting a file"));
+               goto done;
+       }
+       fill_stat_data(&base, &st);
+       fputc('.', stderr);
+
+       avoid_racy();
+       xunlink("new-dir/new");
+       xrmdir("new-dir");
+       xstat_mtime_dir(&st);
+       if (!match_stat_data(&base, &st)) {
+               fputc('\n', stderr);
+               fprintf_ln(stderr, _("directory stat info does not "
+                                    "change after deleting a directory"));
+               goto done;
+       }
+
+       if (rmdir(mtime_dir.buf))
+               die_errno(_("failed to delete directory %s"), mtime_dir.buf);
+       fprintf_ln(stderr, _(" OK"));
+       ret = 1;
+
+done:
+       strbuf_release(&mtime_dir);
+       return ret;
+}
+
 static int mark_ce_flags(const char *path, int flag, int mark)
 {
        int namelen = strlen(path);
@@ -741,6 +902,7 @@ static int reupdate_callback(struct parse_opt_ctx_t *ctx,
 int cmd_update_index(int argc, const char **argv, const char *prefix)
 {
        int newfd, entries, has_errors = 0, line_termination = '\n';
+       int untracked_cache = -1;
        int read_from_stdin = 0;
        int prefix_length = prefix ? strlen(prefix) : 0;
        int preferred_index_format = 0;
@@ -832,6 +994,10 @@ int cmd_update_index(int argc, const char **argv, const char *prefix)
                        N_("write index in this format")),
                OPT_BOOL(0, "split-index", &split_index,
                        N_("enable or disable split index")),
+               OPT_BOOL(0, "untracked-cache", &untracked_cache,
+                       N_("enable/disable untracked cache")),
+               OPT_SET_INT(0, "force-untracked-cache", &untracked_cache,
+                           N_("enable untracked cache without testing the filesystem"), 2),
                OPT_END()
        };
 
@@ -938,6 +1104,28 @@ int cmd_update_index(int argc, const char **argv, const char *prefix)
                the_index.split_index = NULL;
                the_index.cache_changed |= SOMETHING_CHANGED;
        }
+       if (untracked_cache > 0) {
+               struct untracked_cache *uc;
+
+               if (untracked_cache < 2) {
+                       setup_work_tree();
+                       if (!test_if_untracked_cache_is_supported())
+                               return 1;
+               }
+               if (!the_index.untracked) {
+                       uc = xcalloc(1, sizeof(*uc));
+                       strbuf_init(&uc->ident, 100);
+                       uc->exclude_per_dir = ".gitignore";
+                       /* should be the same flags used by git-status */
+                       uc->dir_flags = DIR_SHOW_OTHER_DIRECTORIES | DIR_HIDE_EMPTY_DIRECTORIES;
+                       the_index.untracked = uc;
+               }
+               add_untracked_ident(the_index.untracked);
+               the_index.cache_changed |= UNTRACKED_CHANGED;
+       } else if (!untracked_cache && the_index.untracked) {
+               the_index.untracked = NULL;
+               the_index.cache_changed |= UNTRACKED_CHANGED;
+       }
 
        if (active_cache_changed) {
                if (newfd < 0) {
diff --git a/cache.h b/cache.h
index 1f4226be1580e368b22d62a6e27aa55d37a4dbd7..9da9784824dc3a93db2c753613bb56866a356070 100644 (file)
--- a/cache.h
+++ b/cache.h
@@ -297,8 +297,11 @@ static inline unsigned int canon_mode(unsigned int mode)
 #define RESOLVE_UNDO_CHANGED   (1 << 4)
 #define CACHE_TREE_CHANGED     (1 << 5)
 #define SPLIT_INDEX_ORDERED    (1 << 6)
+#define UNTRACKED_CHANGED      (1 << 7)
 
 struct split_index;
+struct untracked_cache;
+
 struct index_state {
        struct cache_entry **cache;
        unsigned int version;
@@ -312,6 +315,7 @@ struct index_state {
        struct hashmap name_hash;
        struct hashmap dir_hash;
        unsigned char sha1[20];
+       struct untracked_cache *untracked;
 };
 
 extern struct index_state the_index;
@@ -563,6 +567,8 @@ extern void fill_stat_data(struct stat_data *sd, struct stat *st);
  * INODE_CHANGED, and DATA_CHANGED.
  */
 extern int match_stat_data(const struct stat_data *sd, struct stat *st);
+extern int match_stat_data_racy(const struct index_state *istate,
+                               const struct stat_data *sd, struct stat *st);
 
 extern void fill_stat_cache_info(struct cache_entry *ce, struct stat *st);
 
index 70f3191a4f19f10a156d1f2c054943d5147ab049..496e6f8bb0217c40450c61e88fbf1e08fdb3f704 100644 (file)
@@ -2128,3 +2128,14 @@ void mingw_startup()
        /* initialize Unicode console */
        winansi_init();
 }
+
+int uname(struct utsname *buf)
+{
+       DWORD v = GetVersion();
+       memset(buf, 0, sizeof(*buf));
+       strcpy(buf->sysname, "Windows");
+       sprintf(buf->release, "%u.%u", v & 0xff, (v >> 8) & 0xff);
+       /* assuming NT variants only.. */
+       sprintf(buf->version, "%u", (v >> 16) & 0x7fff);
+       return 0;
+}
index 98c5e44294cd1f41d5b177b8d4be5219066c2dcd..738865c6c068ed7d8849aff5a9b533dfb1ef8bab 100644 (file)
@@ -76,6 +76,14 @@ struct itimerval {
 };
 #define ITIMER_REAL 0
 
+struct utsname {
+       char sysname[16];
+       char nodename[1];
+       char release[16];
+       char version[16];
+       char machine[1];
+};
+
 /*
  * sanitize preprocessor namespace polluted by Windows headers defining
  * macros which collide with git local versions
@@ -175,6 +183,7 @@ struct passwd *getpwuid(uid_t uid);
 int setitimer(int type, struct itimerval *in, struct itimerval *out);
 int sigaction(int sig, struct sigaction *in, struct sigaction *out);
 int link(const char *oldpath, const char *newpath);
+int uname(struct utsname *buf);
 
 /*
  * replacements of existing functions
diff --git a/dir.c b/dir.c
index 4183acc082671f135fe64cbcaa66ed3b17bc6364..d318ffcb2a6a51feca5aee993049fd0fa64acc8f 100644 (file)
--- a/dir.c
+++ b/dir.c
@@ -13,6 +13,8 @@
 #include "wildmatch.h"
 #include "pathspec.h"
 #include "utf8.h"
+#include "varint.h"
+#include "ewah/ewok.h"
 
 struct path_simplify {
        int len;
@@ -32,8 +34,22 @@ enum path_treatment {
        path_untracked
 };
 
+/*
+ * Support data structure for our opendir/readdir/closedir wrappers
+ */
+struct cached_dir {
+       DIR *fdir;
+       struct untracked_cache_dir *untracked;
+       int nr_files;
+       int nr_dirs;
+
+       struct dirent *de;
+       const char *file;
+       struct untracked_cache_dir *ucd;
+};
+
 static enum path_treatment read_directory_recursive(struct dir_struct *dir,
-       const char *path, int len,
+       const char *path, int len, struct untracked_cache_dir *untracked,
        int check_only, const struct path_simplify *simplify);
 static int get_dtype(struct dirent *de, const char *path, int len);
 
@@ -510,7 +526,8 @@ void add_exclude(const char *string, const char *base,
        x->el = el;
 }
 
-static void *read_skip_worktree_file_from_index(const char *path, size_t *size)
+static void *read_skip_worktree_file_from_index(const char *path, size_t *size,
+                                               struct sha1_stat *sha1_stat)
 {
        int pos, len;
        unsigned long sz;
@@ -529,6 +546,10 @@ static void *read_skip_worktree_file_from_index(const char *path, size_t *size)
                return NULL;
        }
        *size = xsize_t(sz);
+       if (sha1_stat) {
+               memset(&sha1_stat->stat, 0, sizeof(sha1_stat->stat));
+               hashcpy(sha1_stat->sha1, active_cache[pos]->sha1);
+       }
        return data;
 }
 
@@ -573,11 +594,93 @@ static void trim_trailing_spaces(char *buf)
                *last_space = '\0';
 }
 
-int add_excludes_from_file_to_list(const char *fname,
-                                  const char *base,
-                                  int baselen,
-                                  struct exclude_list *el,
-                                  int check_index)
+/*
+ * Given a subdirectory name and "dir" of the current directory,
+ * search the subdir in "dir" and return it, or create a new one if it
+ * does not exist in "dir".
+ *
+ * If "name" has the trailing slash, it'll be excluded in the search.
+ */
+static struct untracked_cache_dir *lookup_untracked(struct untracked_cache *uc,
+                                                   struct untracked_cache_dir *dir,
+                                                   const char *name, int len)
+{
+       int first, last;
+       struct untracked_cache_dir *d;
+       if (!dir)
+               return NULL;
+       if (len && name[len - 1] == '/')
+               len--;
+       first = 0;
+       last = dir->dirs_nr;
+       while (last > first) {
+               int cmp, next = (last + first) >> 1;
+               d = dir->dirs[next];
+               cmp = strncmp(name, d->name, len);
+               if (!cmp && strlen(d->name) > len)
+                       cmp = -1;
+               if (!cmp)
+                       return d;
+               if (cmp < 0) {
+                       last = next;
+                       continue;
+               }
+               first = next+1;
+       }
+
+       uc->dir_created++;
+       d = xmalloc(sizeof(*d) + len + 1);
+       memset(d, 0, sizeof(*d));
+       memcpy(d->name, name, len);
+       d->name[len] = '\0';
+
+       ALLOC_GROW(dir->dirs, dir->dirs_nr + 1, dir->dirs_alloc);
+       memmove(dir->dirs + first + 1, dir->dirs + first,
+               (dir->dirs_nr - first) * sizeof(*dir->dirs));
+       dir->dirs_nr++;
+       dir->dirs[first] = d;
+       return d;
+}
+
+static void do_invalidate_gitignore(struct untracked_cache_dir *dir)
+{
+       int i;
+       dir->valid = 0;
+       dir->untracked_nr = 0;
+       for (i = 0; i < dir->dirs_nr; i++)
+               do_invalidate_gitignore(dir->dirs[i]);
+}
+
+static void invalidate_gitignore(struct untracked_cache *uc,
+                                struct untracked_cache_dir *dir)
+{
+       uc->gitignore_invalidated++;
+       do_invalidate_gitignore(dir);
+}
+
+static void invalidate_directory(struct untracked_cache *uc,
+                                struct untracked_cache_dir *dir)
+{
+       int i;
+       uc->dir_invalidated++;
+       dir->valid = 0;
+       dir->untracked_nr = 0;
+       for (i = 0; i < dir->dirs_nr; i++)
+               dir->dirs[i]->recurse = 0;
+}
+
+/*
+ * Given a file with name "fname", read it (either from disk, or from
+ * the index if "check_index" is non-zero), parse it and store the
+ * exclude rules in "el".
+ *
+ * If "ss" is not NULL, compute SHA-1 of the exclude file and fill
+ * stat data from disk (only valid if add_excludes returns zero). If
+ * ss_valid is non-zero, "ss" must contain good value as input.
+ */
+static int add_excludes(const char *fname, const char *base, int baselen,
+                       struct exclude_list *el, int check_index,
+                       struct sha1_stat *sha1_stat)
 {
        struct stat st;
        int fd, i, lineno = 1;
@@ -591,7 +694,7 @@ int add_excludes_from_file_to_list(const char *fname,
                if (0 <= fd)
                        close(fd);
                if (!check_index ||
-                   (buf = read_skip_worktree_file_from_index(fname, &size)) == NULL)
+                   (buf = read_skip_worktree_file_from_index(fname, &size, sha1_stat)) == NULL)
                        return -1;
                if (size == 0) {
                        free(buf);
@@ -604,6 +707,11 @@ int add_excludes_from_file_to_list(const char *fname,
        } else {
                size = xsize_t(st.st_size);
                if (size == 0) {
+                       if (sha1_stat) {
+                               fill_stat_data(&sha1_stat->stat, &st);
+                               hashcpy(sha1_stat->sha1, EMPTY_BLOB_SHA1_BIN);
+                               sha1_stat->valid = 1;
+                       }
                        close(fd);
                        return 0;
                }
@@ -615,6 +723,22 @@ int add_excludes_from_file_to_list(const char *fname,
                }
                buf[size++] = '\n';
                close(fd);
+               if (sha1_stat) {
+                       int pos;
+                       if (sha1_stat->valid &&
+                           !match_stat_data_racy(&the_index, &sha1_stat->stat, &st))
+                               ; /* no content change, ss->sha1 still good */
+                       else if (check_index &&
+                                (pos = cache_name_pos(fname, strlen(fname))) >= 0 &&
+                                !ce_stage(active_cache[pos]) &&
+                                ce_uptodate(active_cache[pos]) &&
+                                !would_convert_to_git(fname))
+                               hashcpy(sha1_stat->sha1, active_cache[pos]->sha1);
+                       else
+                               hash_sha1_file(buf, size, "blob", sha1_stat->sha1);
+                       fill_stat_data(&sha1_stat->stat, &st);
+                       sha1_stat->valid = 1;
+               }
        }
 
        el->filebuf = buf;
@@ -638,6 +762,13 @@ int add_excludes_from_file_to_list(const char *fname,
        return 0;
 }
 
+int add_excludes_from_file_to_list(const char *fname, const char *base,
+                                  int baselen, struct exclude_list *el,
+                                  int check_index)
+{
+       return add_excludes(fname, base, baselen, el, check_index, NULL);
+}
+
 struct exclude_list *add_exclude_list(struct dir_struct *dir,
                                      int group_type, const char *src)
 {
@@ -655,14 +786,28 @@ struct exclude_list *add_exclude_list(struct dir_struct *dir,
 /*
  * Used to set up core.excludesfile and .git/info/exclude lists.
  */
-void add_excludes_from_file(struct dir_struct *dir, const char *fname)
+static void add_excludes_from_file_1(struct dir_struct *dir, const char *fname,
+                                    struct sha1_stat *sha1_stat)
 {
        struct exclude_list *el;
+       /*
+        * catch setup_standard_excludes() that's called before
+        * dir->untracked is assigned. That function behaves
+        * differently when dir->untracked is non-NULL.
+        */
+       if (!dir->untracked)
+               dir->unmanaged_exclude_files++;
        el = add_exclude_list(dir, EXC_FILE, fname);
-       if (add_excludes_from_file_to_list(fname, "", 0, el, 0) < 0)
+       if (add_excludes(fname, "", 0, el, 0, sha1_stat) < 0)
                die("cannot use %s as an exclude file", fname);
 }
 
+void add_excludes_from_file(struct dir_struct *dir, const char *fname)
+{
+       dir->unmanaged_exclude_files++; /* see validate_untracked_cache() */
+       add_excludes_from_file_1(dir, fname, NULL);
+}
+
 int match_basename(const char *basename, int basenamelen,
                   const char *pattern, int prefix, int patternlen,
                   int flags)
@@ -837,6 +982,7 @@ static void prep_exclude(struct dir_struct *dir, const char *base, int baselen)
        struct exclude_list_group *group;
        struct exclude_list *el;
        struct exclude_stack *stk = NULL;
+       struct untracked_cache_dir *untracked;
        int current;
 
        group = &dir->exclude_list_group[EXC_DIRS];
@@ -874,8 +1020,14 @@ static void prep_exclude(struct dir_struct *dir, const char *base, int baselen)
        /* Read from the parent directories and push them down. */
        current = stk ? stk->baselen : -1;
        strbuf_setlen(&dir->basebuf, current < 0 ? 0 : current);
+       if (dir->untracked)
+               untracked = stk ? stk->ucd : dir->untracked->root;
+       else
+               untracked = NULL;
+
        while (current < baselen) {
                const char *cp;
+               struct sha1_stat sha1_stat;
 
                stk = xcalloc(1, sizeof(*stk));
                if (current < 0) {
@@ -886,10 +1038,15 @@ static void prep_exclude(struct dir_struct *dir, const char *base, int baselen)
                        if (!cp)
                                die("oops in prep_exclude");
                        cp++;
+                       untracked =
+                               lookup_untracked(dir->untracked, untracked,
+                                                base + current,
+                                                cp - base - current);
                }
                stk->prev = dir->exclude_stack;
                stk->baselen = cp - base;
                stk->exclude_ix = group->nr;
+               stk->ucd = untracked;
                el = add_exclude_list(dir, EXC_DIRS, NULL);
                strbuf_add(&dir->basebuf, base + current, stk->baselen - current);
                assert(stk->baselen == dir->basebuf.len);
@@ -912,7 +1069,23 @@ static void prep_exclude(struct dir_struct *dir, const char *base, int baselen)
                }
 
                /* Try to read per-directory file */
-               if (dir->exclude_per_dir) {
+               hashclr(sha1_stat.sha1);
+               sha1_stat.valid = 0;
+               if (dir->exclude_per_dir &&
+                   /*
+                    * If we know that no files have been added in
+                    * this directory (i.e. valid_cached_dir() has
+                    * been executed and set untracked->valid) ..
+                    */
+                   (!untracked || !untracked->valid ||
+                    /*
+                     * .. and .gitignore does not exist before
+                     * (i.e. null exclude_sha1 and skip_worktree is
+                     * not set). Then we can skip loading .gitignore,
+                     * which would result in ENOENT anyway.
+                     * skip_worktree is taken care in read_directory()
+                     */
+                    !is_null_sha1(untracked->exclude_sha1))) {
                        /*
                         * dir->basebuf gets reused by the traversal, but we
                         * need fname to remain unchanged to ensure the src
@@ -925,8 +1098,27 @@ static void prep_exclude(struct dir_struct *dir, const char *base, int baselen)
                        strbuf_addbuf(&sb, &dir->basebuf);
                        strbuf_addstr(&sb, dir->exclude_per_dir);
                        el->src = strbuf_detach(&sb, NULL);
-                       add_excludes_from_file_to_list(el->src, el->src,
-                                                      stk->baselen, el, 1);
+                       add_excludes(el->src, el->src, stk->baselen, el, 1,
+                                    untracked ? &sha1_stat : NULL);
+               }
+               /*
+                * NEEDSWORK: when untracked cache is enabled, prep_exclude()
+                * will first be called in valid_cached_dir() then maybe many
+                * times more in last_exclude_matching(). When the cache is
+                * used, last_exclude_matching() will not be called and
+                * reading .gitignore content will be a waste.
+                *
+                * So when it's called by valid_cached_dir() and we can get
+                * .gitignore SHA-1 from the index (i.e. .gitignore is not
+                * modified on work tree), we could delay reading the
+                * .gitignore content until we absolutely need it in
+                * last_exclude_matching(). Be careful about ignore rule
+                * order, though, if you do that.
+                */
+               if (untracked &&
+                   hashcmp(sha1_stat.sha1, untracked->exclude_sha1)) {
+                       invalidate_gitignore(dir->untracked, untracked);
+                       hashcpy(untracked->exclude_sha1, sha1_stat.sha1);
                }
                dir->exclude_stack = stk;
                current = stk->baselen;
@@ -1107,6 +1299,7 @@ static enum exist_status directory_exists_in_index(const char *dirname, int len)
  *  (c) otherwise, we recurse into it.
  */
 static enum path_treatment treat_directory(struct dir_struct *dir,
+       struct untracked_cache_dir *untracked,
        const char *dirname, int len, int exclude,
        const struct path_simplify *simplify)
 {
@@ -1134,7 +1327,9 @@ static enum path_treatment treat_directory(struct dir_struct *dir,
        if (!(dir->flags & DIR_HIDE_EMPTY_DIRECTORIES))
                return exclude ? path_excluded : path_untracked;
 
-       return read_directory_recursive(dir, dirname, len, 1, simplify);
+       untracked = lookup_untracked(dir->untracked, untracked, dirname, len);
+       return read_directory_recursive(dir, dirname, len,
+                                       untracked, 1, simplify);
 }
 
 /*
@@ -1250,6 +1445,7 @@ static int get_dtype(struct dirent *de, const char *path, int len)
 }
 
 static enum path_treatment treat_one_path(struct dir_struct *dir,
+                                         struct untracked_cache_dir *untracked,
                                          struct strbuf *path,
                                          const struct path_simplify *simplify,
                                          int dtype, struct dirent *de)
@@ -1302,7 +1498,7 @@ static enum path_treatment treat_one_path(struct dir_struct *dir,
                return path_none;
        case DT_DIR:
                strbuf_addch(path, '/');
-               return treat_directory(dir, path->buf, path->len, exclude,
+               return treat_directory(dir, untracked, path->buf, path->len, exclude,
                        simplify);
        case DT_REG:
        case DT_LNK:
@@ -1310,14 +1506,52 @@ static enum path_treatment treat_one_path(struct dir_struct *dir,
        }
 }
 
+static enum path_treatment treat_path_fast(struct dir_struct *dir,
+                                          struct untracked_cache_dir *untracked,
+                                          struct cached_dir *cdir,
+                                          struct strbuf *path,
+                                          int baselen,
+                                          const struct path_simplify *simplify)
+{
+       strbuf_setlen(path, baselen);
+       if (!cdir->ucd) {
+               strbuf_addstr(path, cdir->file);
+               return path_untracked;
+       }
+       strbuf_addstr(path, cdir->ucd->name);
+       /* treat_one_path() does this before it calls treat_directory() */
+       if (path->buf[path->len - 1] != '/')
+               strbuf_addch(path, '/');
+       if (cdir->ucd->check_only)
+               /*
+                * check_only is set as a result of treat_directory() getting
+                * to its bottom. Verify again the same set of directories
+                * with check_only set.
+                */
+               return read_directory_recursive(dir, path->buf, path->len,
+                                               cdir->ucd, 1, simplify);
+       /*
+        * We get path_recurse in the first run when
+        * directory_exists_in_index() returns index_nonexistent. We
+        * are sure that new changes in the index does not impact the
+        * outcome. Return now.
+        */
+       return path_recurse;
+}
+
 static enum path_treatment treat_path(struct dir_struct *dir,
-                                     struct dirent *de,
+                                     struct untracked_cache_dir *untracked,
+                                     struct cached_dir *cdir,
                                      struct strbuf *path,
                                      int baselen,
                                      const struct path_simplify *simplify)
 {
        int dtype;
+       struct dirent *de = cdir->de;
 
+       if (!de)
+               return treat_path_fast(dir, untracked, cdir, path,
+                                      baselen, simplify);
        if (is_dot_or_dotdot(de->d_name) || !strcmp(de->d_name, ".git"))
                return path_none;
        strbuf_setlen(path, baselen);
@@ -1326,7 +1560,121 @@ static enum path_treatment treat_path(struct dir_struct *dir,
                return path_none;
 
        dtype = DTYPE(de);
-       return treat_one_path(dir, path, simplify, dtype, de);
+       return treat_one_path(dir, untracked, path, simplify, dtype, de);
+}
+
+static void add_untracked(struct untracked_cache_dir *dir, const char *name)
+{
+       if (!dir)
+               return;
+       ALLOC_GROW(dir->untracked, dir->untracked_nr + 1,
+                  dir->untracked_alloc);
+       dir->untracked[dir->untracked_nr++] = xstrdup(name);
+}
+
+static int valid_cached_dir(struct dir_struct *dir,
+                           struct untracked_cache_dir *untracked,
+                           struct strbuf *path,
+                           int check_only)
+{
+       struct stat st;
+
+       if (!untracked)
+               return 0;
+
+       if (stat(path->len ? path->buf : ".", &st)) {
+               invalidate_directory(dir->untracked, untracked);
+               memset(&untracked->stat_data, 0, sizeof(untracked->stat_data));
+               return 0;
+       }
+       if (!untracked->valid ||
+           match_stat_data_racy(&the_index, &untracked->stat_data, &st)) {
+               if (untracked->valid)
+                       invalidate_directory(dir->untracked, untracked);
+               fill_stat_data(&untracked->stat_data, &st);
+               return 0;
+       }
+
+       if (untracked->check_only != !!check_only) {
+               invalidate_directory(dir->untracked, untracked);
+               return 0;
+       }
+
+       /*
+        * prep_exclude will be called eventually on this directory,
+        * but it's called much later in last_exclude_matching(). We
+        * need it now to determine the validity of the cache for this
+        * path. The next calls will be nearly no-op, the way
+        * prep_exclude() is designed.
+        */
+       if (path->len && path->buf[path->len - 1] != '/') {
+               strbuf_addch(path, '/');
+               prep_exclude(dir, path->buf, path->len);
+               strbuf_setlen(path, path->len - 1);
+       } else
+               prep_exclude(dir, path->buf, path->len);
+
+       /* hopefully prep_exclude() haven't invalidated this entry... */
+       return untracked->valid;
+}
+
+static int open_cached_dir(struct cached_dir *cdir,
+                          struct dir_struct *dir,
+                          struct untracked_cache_dir *untracked,
+                          struct strbuf *path,
+                          int check_only)
+{
+       memset(cdir, 0, sizeof(*cdir));
+       cdir->untracked = untracked;
+       if (valid_cached_dir(dir, untracked, path, check_only))
+               return 0;
+       cdir->fdir = opendir(path->len ? path->buf : ".");
+       if (dir->untracked)
+               dir->untracked->dir_opened++;
+       if (!cdir->fdir)
+               return -1;
+       return 0;
+}
+
+static int read_cached_dir(struct cached_dir *cdir)
+{
+       if (cdir->fdir) {
+               cdir->de = readdir(cdir->fdir);
+               if (!cdir->de)
+                       return -1;
+               return 0;
+       }
+       while (cdir->nr_dirs < cdir->untracked->dirs_nr) {
+               struct untracked_cache_dir *d = cdir->untracked->dirs[cdir->nr_dirs];
+               if (!d->recurse) {
+                       cdir->nr_dirs++;
+                       continue;
+               }
+               cdir->ucd = d;
+               cdir->nr_dirs++;
+               return 0;
+       }
+       cdir->ucd = NULL;
+       if (cdir->nr_files < cdir->untracked->untracked_nr) {
+               struct untracked_cache_dir *d = cdir->untracked;
+               cdir->file = d->untracked[cdir->nr_files++];
+               return 0;
+       }
+       return -1;
+}
+
+static void close_cached_dir(struct cached_dir *cdir)
+{
+       if (cdir->fdir)
+               closedir(cdir->fdir);
+       /*
+        * We have gone through this directory and found no untracked
+        * entries. Mark it valid.
+        */
+       if (cdir->untracked) {
+               cdir->untracked->valid = 1;
+               cdir->untracked->recurse = 1;
+       }
 }
 
 /*
@@ -1342,38 +1690,48 @@ static enum path_treatment treat_path(struct dir_struct *dir,
  */
 static enum path_treatment read_directory_recursive(struct dir_struct *dir,
                                    const char *base, int baselen,
-                                   int check_only,
+                                   struct untracked_cache_dir *untracked, int check_only,
                                    const struct path_simplify *simplify)
 {
-       DIR *fdir;
+       struct cached_dir cdir;
        enum path_treatment state, subdir_state, dir_state = path_none;
-       struct dirent *de;
        struct strbuf path = STRBUF_INIT;
 
        strbuf_add(&path, base, baselen);
 
-       fdir = opendir(path.len ? path.buf : ".");
-       if (!fdir)
+       if (open_cached_dir(&cdir, dir, untracked, &path, check_only))
                goto out;
 
-       while ((de = readdir(fdir)) != NULL) {
+       if (untracked)
+               untracked->check_only = !!check_only;
+
+       while (!read_cached_dir(&cdir)) {
                /* check how the file or directory should be treated */
-               state = treat_path(dir, de, &path, baselen, simplify);
+               state = treat_path(dir, untracked, &cdir, &path, baselen, simplify);
+
                if (state > dir_state)
                        dir_state = state;
 
                /* recurse into subdir if instructed by treat_path */
                if (state == path_recurse) {
-                       subdir_state = read_directory_recursive(dir, path.buf,
-                               path.len, check_only, simplify);
+                       struct untracked_cache_dir *ud;
+                       ud = lookup_untracked(dir->untracked, untracked,
+                                             path.buf + baselen,
+                                             path.len - baselen);
+                       subdir_state =
+                               read_directory_recursive(dir, path.buf, path.len,
+                                                        ud, check_only, simplify);
                        if (subdir_state > dir_state)
                                dir_state = subdir_state;
                }
 
                if (check_only) {
                        /* abort early if maximum state has been reached */
-                       if (dir_state == path_untracked)
+                       if (dir_state == path_untracked) {
+                               if (cdir.fdir)
+                                       add_untracked(untracked, path.buf + baselen);
                                break;
+                       }
                        /* skip the dir_add_* part */
                        continue;
                }
@@ -1391,15 +1749,18 @@ static enum path_treatment read_directory_recursive(struct dir_struct *dir,
                        break;
 
                case path_untracked:
-                       if (!(dir->flags & DIR_SHOW_IGNORED))
-                               dir_add_name(dir, path.buf, path.len);
+                       if (dir->flags & DIR_SHOW_IGNORED)
+                               break;
+                       dir_add_name(dir, path.buf, path.len);
+                       if (cdir.fdir)
+                               add_untracked(untracked, path.buf + baselen);
                        break;
 
                default:
                        break;
                }
        }
-       closedir(fdir);
+       close_cached_dir(&cdir);
  out:
        strbuf_release(&path);
 
@@ -1469,7 +1830,7 @@ static int treat_leading_path(struct dir_struct *dir,
                        break;
                if (simplify_away(sb.buf, sb.len, simplify))
                        break;
-               if (treat_one_path(dir, &sb, simplify,
+               if (treat_one_path(dir, NULL, &sb, simplify,
                                   DT_DIR, NULL) == path_none)
                        break; /* do not recurse into it */
                if (len <= baselen) {
@@ -1482,9 +1843,139 @@ static int treat_leading_path(struct dir_struct *dir,
        return rc;
 }
 
+static const char *get_ident_string(void)
+{
+       static struct strbuf sb = STRBUF_INIT;
+       struct utsname uts;
+
+       if (sb.len)
+               return sb.buf;
+       if (uname(&uts))
+               die_errno(_("failed to get kernel name and information"));
+       strbuf_addf(&sb, "Location %s, system %s %s %s", get_git_work_tree(),
+                   uts.sysname, uts.release, uts.version);
+       return sb.buf;
+}
+
+static int ident_in_untracked(const struct untracked_cache *uc)
+{
+       const char *end = uc->ident.buf + uc->ident.len;
+       const char *p   = uc->ident.buf;
+
+       for (p = uc->ident.buf; p < end; p += strlen(p) + 1)
+               if (!strcmp(p, get_ident_string()))
+                       return 1;
+       return 0;
+}
+
+void add_untracked_ident(struct untracked_cache *uc)
+{
+       if (ident_in_untracked(uc))
+               return;
+       strbuf_addstr(&uc->ident, get_ident_string());
+       /* this strbuf contains a list of strings, save NUL too */
+       strbuf_addch(&uc->ident, 0);
+}
+
+static struct untracked_cache_dir *validate_untracked_cache(struct dir_struct *dir,
+                                                     int base_len,
+                                                     const struct pathspec *pathspec)
+{
+       struct untracked_cache_dir *root;
+       int i;
+
+       if (!dir->untracked || getenv("GIT_DISABLE_UNTRACKED_CACHE"))
+               return NULL;
+
+       /*
+        * We only support $GIT_DIR/info/exclude and core.excludesfile
+        * as the global ignore rule files. Any other additions
+        * (e.g. from command line) invalidate the cache. This
+        * condition also catches running setup_standard_excludes()
+        * before setting dir->untracked!
+        */
+       if (dir->unmanaged_exclude_files)
+               return NULL;
+
+       /*
+        * Optimize for the main use case only: whole-tree git
+        * status. More work involved in treat_leading_path() if we
+        * use cache on just a subset of the worktree. pathspec
+        * support could make the matter even worse.
+        */
+       if (base_len || (pathspec && pathspec->nr))
+               return NULL;
+
+       /* Different set of flags may produce different results */
+       if (dir->flags != dir->untracked->dir_flags ||
+           /*
+            * See treat_directory(), case index_nonexistent. Without
+            * this flag, we may need to also cache .git file content
+            * for the resolve_gitlink_ref() call, which we don't.
+            */
+           !(dir->flags & DIR_SHOW_OTHER_DIRECTORIES) ||
+           /* We don't support collecting ignore files */
+           (dir->flags & (DIR_SHOW_IGNORED | DIR_SHOW_IGNORED_TOO |
+                          DIR_COLLECT_IGNORED)))
+               return NULL;
+
+       /*
+        * If we use .gitignore in the cache and now you change it to
+        * .gitexclude, everything will go wrong.
+        */
+       if (dir->exclude_per_dir != dir->untracked->exclude_per_dir &&
+           strcmp(dir->exclude_per_dir, dir->untracked->exclude_per_dir))
+               return NULL;
+
+       /*
+        * EXC_CMDL is not considered in the cache. If people set it,
+        * skip the cache.
+        */
+       if (dir->exclude_list_group[EXC_CMDL].nr)
+               return NULL;
+
+       /*
+        * An optimization in prep_exclude() does not play well with
+        * CE_SKIP_WORKTREE. It's a rare case anyway, if a single
+        * entry has that bit set, disable the whole untracked cache.
+        */
+       for (i = 0; i < active_nr; i++)
+               if (ce_skip_worktree(active_cache[i]))
+                       return NULL;
+
+       if (!ident_in_untracked(dir->untracked)) {
+               warning(_("Untracked cache is disabled on this system."));
+               return NULL;
+       }
+
+       if (!dir->untracked->root) {
+               const int len = sizeof(*dir->untracked->root);
+               dir->untracked->root = xmalloc(len);
+               memset(dir->untracked->root, 0, len);
+       }
+
+       /* Validate $GIT_DIR/info/exclude and core.excludesfile */
+       root = dir->untracked->root;
+       if (hashcmp(dir->ss_info_exclude.sha1,
+                   dir->untracked->ss_info_exclude.sha1)) {
+               invalidate_gitignore(dir->untracked, root);
+               dir->untracked->ss_info_exclude = dir->ss_info_exclude;
+       }
+       if (hashcmp(dir->ss_excludes_file.sha1,
+                   dir->untracked->ss_excludes_file.sha1)) {
+               invalidate_gitignore(dir->untracked, root);
+               dir->untracked->ss_excludes_file = dir->ss_excludes_file;
+       }
+
+       /* Make sure this directory is not dropped out at saving phase */
+       root->recurse = 1;
+       return root;
+}
+
 int read_directory(struct dir_struct *dir, const char *path, int len, const struct pathspec *pathspec)
 {
        struct path_simplify *simplify;
+       struct untracked_cache_dir *untracked;
 
        /*
         * Check out create_simplify()
@@ -1508,11 +1999,39 @@ int read_directory(struct dir_struct *dir, const char *path, int len, const stru
         * create_simplify().
         */
        simplify = create_simplify(pathspec ? pathspec->_raw : NULL);
+       untracked = validate_untracked_cache(dir, len, pathspec);
+       if (!untracked)
+               /*
+                * make sure untracked cache code path is disabled,
+                * e.g. prep_exclude()
+                */
+               dir->untracked = NULL;
        if (!len || treat_leading_path(dir, path, len, simplify))
-               read_directory_recursive(dir, path, len, 0, simplify);
+               read_directory_recursive(dir, path, len, untracked, 0, simplify);
        free_simplify(simplify);
        qsort(dir->entries, dir->nr, sizeof(struct dir_entry *), cmp_name);
        qsort(dir->ignored, dir->ignored_nr, sizeof(struct dir_entry *), cmp_name);
+       if (dir->untracked) {
+               static struct trace_key trace_untracked_stats = TRACE_KEY_INIT(UNTRACKED_STATS);
+               trace_printf_key(&trace_untracked_stats,
+                                "node creation: %u\n"
+                                "gitignore invalidation: %u\n"
+                                "directory invalidation: %u\n"
+                                "opendir: %u\n",
+                                dir->untracked->dir_created,
+                                dir->untracked->gitignore_invalidated,
+                                dir->untracked->dir_invalidated,
+                                dir->untracked->dir_opened);
+               if (dir->untracked == the_index.untracked &&
+                   (dir->untracked->dir_opened ||
+                    dir->untracked->gitignore_invalidated ||
+                    dir->untracked->dir_invalidated))
+                       the_index.cache_changed |= UNTRACKED_CHANGED;
+               if (dir->untracked != the_index.untracked) {
+                       free(dir->untracked);
+                       dir->untracked = NULL;
+               }
+       }
        return dir->nr;
 }
 
@@ -1678,12 +2197,14 @@ void setup_standard_excludes(struct dir_struct *dir)
        if (!excludes_file)
                excludes_file = xdg_config_home("ignore");
        if (excludes_file && !access_or_warn(excludes_file, R_OK, 0))
-               add_excludes_from_file(dir, excludes_file);
+               add_excludes_from_file_1(dir, excludes_file,
+                                        dir->untracked ? &dir->ss_excludes_file : NULL);
 
        /* per repository user preference */
        path = git_path("info/exclude");
        if (!access_or_warn(path, R_OK, 0))
-               add_excludes_from_file(dir, path);
+               add_excludes_from_file_1(dir, path,
+                                        dir->untracked ? &dir->ss_info_exclude : NULL);
 }
 
 int remove_path(const char *name)
@@ -1735,3 +2256,404 @@ void clear_directory(struct dir_struct *dir)
        }
        strbuf_release(&dir->basebuf);
 }
+
+struct ondisk_untracked_cache {
+       struct stat_data info_exclude_stat;
+       struct stat_data excludes_file_stat;
+       uint32_t dir_flags;
+       unsigned char info_exclude_sha1[20];
+       unsigned char excludes_file_sha1[20];
+       char exclude_per_dir[FLEX_ARRAY];
+};
+
+#define ouc_size(len) (offsetof(struct ondisk_untracked_cache, exclude_per_dir) + len + 1)
+
+struct write_data {
+       int index;         /* number of written untracked_cache_dir */
+       struct ewah_bitmap *check_only; /* from untracked_cache_dir */
+       struct ewah_bitmap *valid;      /* from untracked_cache_dir */
+       struct ewah_bitmap *sha1_valid; /* set if exclude_sha1 is not null */
+       struct strbuf out;
+       struct strbuf sb_stat;
+       struct strbuf sb_sha1;
+};
+
+static void stat_data_to_disk(struct stat_data *to, const struct stat_data *from)
+{
+       to->sd_ctime.sec  = htonl(from->sd_ctime.sec);
+       to->sd_ctime.nsec = htonl(from->sd_ctime.nsec);
+       to->sd_mtime.sec  = htonl(from->sd_mtime.sec);
+       to->sd_mtime.nsec = htonl(from->sd_mtime.nsec);
+       to->sd_dev        = htonl(from->sd_dev);
+       to->sd_ino        = htonl(from->sd_ino);
+       to->sd_uid        = htonl(from->sd_uid);
+       to->sd_gid        = htonl(from->sd_gid);
+       to->sd_size       = htonl(from->sd_size);
+}
+
+static void write_one_dir(struct untracked_cache_dir *untracked,
+                         struct write_data *wd)
+{
+       struct stat_data stat_data;
+       struct strbuf *out = &wd->out;
+       unsigned char intbuf[16];
+       unsigned int intlen, value;
+       int i = wd->index++;
+
+       /*
+        * untracked_nr should be reset whenever valid is clear, but
+        * for safety..
+        */
+       if (!untracked->valid) {
+               untracked->untracked_nr = 0;
+               untracked->check_only = 0;
+       }
+
+       if (untracked->check_only)
+               ewah_set(wd->check_only, i);
+       if (untracked->valid) {
+               ewah_set(wd->valid, i);
+               stat_data_to_disk(&stat_data, &untracked->stat_data);
+               strbuf_add(&wd->sb_stat, &stat_data, sizeof(stat_data));
+       }
+       if (!is_null_sha1(untracked->exclude_sha1)) {
+               ewah_set(wd->sha1_valid, i);
+               strbuf_add(&wd->sb_sha1, untracked->exclude_sha1, 20);
+       }
+
+       intlen = encode_varint(untracked->untracked_nr, intbuf);
+       strbuf_add(out, intbuf, intlen);
+
+       /* skip non-recurse directories */
+       for (i = 0, value = 0; i < untracked->dirs_nr; i++)
+               if (untracked->dirs[i]->recurse)
+                       value++;
+       intlen = encode_varint(value, intbuf);
+       strbuf_add(out, intbuf, intlen);
+
+       strbuf_add(out, untracked->name, strlen(untracked->name) + 1);
+
+       for (i = 0; i < untracked->untracked_nr; i++)
+               strbuf_add(out, untracked->untracked[i],
+                          strlen(untracked->untracked[i]) + 1);
+
+       for (i = 0; i < untracked->dirs_nr; i++)
+               if (untracked->dirs[i]->recurse)
+                       write_one_dir(untracked->dirs[i], wd);
+}
+
+void write_untracked_extension(struct strbuf *out, struct untracked_cache *untracked)
+{
+       struct ondisk_untracked_cache *ouc;
+       struct write_data wd;
+       unsigned char varbuf[16];
+       int len = 0, varint_len;
+       if (untracked->exclude_per_dir)
+               len = strlen(untracked->exclude_per_dir);
+       ouc = xmalloc(sizeof(*ouc) + len + 1);
+       stat_data_to_disk(&ouc->info_exclude_stat, &untracked->ss_info_exclude.stat);
+       stat_data_to_disk(&ouc->excludes_file_stat, &untracked->ss_excludes_file.stat);
+       hashcpy(ouc->info_exclude_sha1, untracked->ss_info_exclude.sha1);
+       hashcpy(ouc->excludes_file_sha1, untracked->ss_excludes_file.sha1);
+       ouc->dir_flags = htonl(untracked->dir_flags);
+       memcpy(ouc->exclude_per_dir, untracked->exclude_per_dir, len + 1);
+
+       varint_len = encode_varint(untracked->ident.len, varbuf);
+       strbuf_add(out, varbuf, varint_len);
+       strbuf_add(out, untracked->ident.buf, untracked->ident.len);
+
+       strbuf_add(out, ouc, ouc_size(len));
+       free(ouc);
+       ouc = NULL;
+
+       if (!untracked->root) {
+               varint_len = encode_varint(0, varbuf);
+               strbuf_add(out, varbuf, varint_len);
+               return;
+       }
+
+       wd.index      = 0;
+       wd.check_only = ewah_new();
+       wd.valid      = ewah_new();
+       wd.sha1_valid = ewah_new();
+       strbuf_init(&wd.out, 1024);
+       strbuf_init(&wd.sb_stat, 1024);
+       strbuf_init(&wd.sb_sha1, 1024);
+       write_one_dir(untracked->root, &wd);
+
+       varint_len = encode_varint(wd.index, varbuf);
+       strbuf_add(out, varbuf, varint_len);
+       strbuf_addbuf(out, &wd.out);
+       ewah_serialize_strbuf(wd.valid, out);
+       ewah_serialize_strbuf(wd.check_only, out);
+       ewah_serialize_strbuf(wd.sha1_valid, out);
+       strbuf_addbuf(out, &wd.sb_stat);
+       strbuf_addbuf(out, &wd.sb_sha1);
+       strbuf_addch(out, '\0'); /* safe guard for string lists */
+
+       ewah_free(wd.valid);
+       ewah_free(wd.check_only);
+       ewah_free(wd.sha1_valid);
+       strbuf_release(&wd.out);
+       strbuf_release(&wd.sb_stat);
+       strbuf_release(&wd.sb_sha1);
+}
+
+static void free_untracked(struct untracked_cache_dir *ucd)
+{
+       int i;
+       if (!ucd)
+               return;
+       for (i = 0; i < ucd->dirs_nr; i++)
+               free_untracked(ucd->dirs[i]);
+       for (i = 0; i < ucd->untracked_nr; i++)
+               free(ucd->untracked[i]);
+       free(ucd->untracked);
+       free(ucd->dirs);
+       free(ucd);
+}
+
+void free_untracked_cache(struct untracked_cache *uc)
+{
+       if (uc)
+               free_untracked(uc->root);
+       free(uc);
+}
+
+struct read_data {
+       int index;
+       struct untracked_cache_dir **ucd;
+       struct ewah_bitmap *check_only;
+       struct ewah_bitmap *valid;
+       struct ewah_bitmap *sha1_valid;
+       const unsigned char *data;
+       const unsigned char *end;
+};
+
+static void stat_data_from_disk(struct stat_data *to, const struct stat_data *from)
+{
+       to->sd_ctime.sec  = get_be32(&from->sd_ctime.sec);
+       to->sd_ctime.nsec = get_be32(&from->sd_ctime.nsec);
+       to->sd_mtime.sec  = get_be32(&from->sd_mtime.sec);
+       to->sd_mtime.nsec = get_be32(&from->sd_mtime.nsec);
+       to->sd_dev        = get_be32(&from->sd_dev);
+       to->sd_ino        = get_be32(&from->sd_ino);
+       to->sd_uid        = get_be32(&from->sd_uid);
+       to->sd_gid        = get_be32(&from->sd_gid);
+       to->sd_size       = get_be32(&from->sd_size);
+}
+
+static int read_one_dir(struct untracked_cache_dir **untracked_,
+                       struct read_data *rd)
+{
+       struct untracked_cache_dir ud, *untracked;
+       const unsigned char *next, *data = rd->data, *end = rd->end;
+       unsigned int value;
+       int i, len;
+
+       memset(&ud, 0, sizeof(ud));
+
+       next = data;
+       value = decode_varint(&next);
+       if (next > end)
+               return -1;
+       ud.recurse         = 1;
+       ud.untracked_alloc = value;
+       ud.untracked_nr    = value;
+       if (ud.untracked_nr)
+               ud.untracked = xmalloc(sizeof(*ud.untracked) * ud.untracked_nr);
+       data = next;
+
+       next = data;
+       ud.dirs_alloc = ud.dirs_nr = decode_varint(&next);
+       if (next > end)
+               return -1;
+       ud.dirs = xmalloc(sizeof(*ud.dirs) * ud.dirs_nr);
+       data = next;
+
+       len = strlen((const char *)data);
+       next = data + len + 1;
+       if (next > rd->end)
+               return -1;
+       *untracked_ = untracked = xmalloc(sizeof(*untracked) + len);
+       memcpy(untracked, &ud, sizeof(ud));
+       memcpy(untracked->name, data, len + 1);
+       data = next;
+
+       for (i = 0; i < untracked->untracked_nr; i++) {
+               len = strlen((const char *)data);
+               next = data + len + 1;
+               if (next > rd->end)
+                       return -1;
+               untracked->untracked[i] = xstrdup((const char*)data);
+               data = next;
+       }
+
+       rd->ucd[rd->index++] = untracked;
+       rd->data = data;
+
+       for (i = 0; i < untracked->dirs_nr; i++) {
+               len = read_one_dir(untracked->dirs + i, rd);
+               if (len < 0)
+                       return -1;
+       }
+       return 0;
+}
+
+static void set_check_only(size_t pos, void *cb)
+{
+       struct read_data *rd = cb;
+       struct untracked_cache_dir *ud = rd->ucd[pos];
+       ud->check_only = 1;
+}
+
+static void read_stat(size_t pos, void *cb)
+{
+       struct read_data *rd = cb;
+       struct untracked_cache_dir *ud = rd->ucd[pos];
+       if (rd->data + sizeof(struct stat_data) > rd->end) {
+               rd->data = rd->end + 1;
+               return;
+       }
+       stat_data_from_disk(&ud->stat_data, (struct stat_data *)rd->data);
+       rd->data += sizeof(struct stat_data);
+       ud->valid = 1;
+}
+
+static void read_sha1(size_t pos, void *cb)
+{
+       struct read_data *rd = cb;
+       struct untracked_cache_dir *ud = rd->ucd[pos];
+       if (rd->data + 20 > rd->end) {
+               rd->data = rd->end + 1;
+               return;
+       }
+       hashcpy(ud->exclude_sha1, rd->data);
+       rd->data += 20;
+}
+
+static void load_sha1_stat(struct sha1_stat *sha1_stat,
+                          const struct stat_data *stat,
+                          const unsigned char *sha1)
+{
+       stat_data_from_disk(&sha1_stat->stat, stat);
+       hashcpy(sha1_stat->sha1, sha1);
+       sha1_stat->valid = 1;
+}
+
+struct untracked_cache *read_untracked_extension(const void *data, unsigned long sz)
+{
+       const struct ondisk_untracked_cache *ouc;
+       struct untracked_cache *uc;
+       struct read_data rd;
+       const unsigned char *next = data, *end = (const unsigned char *)data + sz;
+       const char *ident;
+       int ident_len, len;
+
+       if (sz <= 1 || end[-1] != '\0')
+               return NULL;
+       end--;
+
+       ident_len = decode_varint(&next);
+       if (next + ident_len > end)
+               return NULL;
+       ident = (const char *)next;
+       next += ident_len;
+
+       ouc = (const struct ondisk_untracked_cache *)next;
+       if (next + ouc_size(0) > end)
+               return NULL;
+
+       uc = xcalloc(1, sizeof(*uc));
+       strbuf_init(&uc->ident, ident_len);
+       strbuf_add(&uc->ident, ident, ident_len);
+       load_sha1_stat(&uc->ss_info_exclude, &ouc->info_exclude_stat,
+                      ouc->info_exclude_sha1);
+       load_sha1_stat(&uc->ss_excludes_file, &ouc->excludes_file_stat,
+                      ouc->excludes_file_sha1);
+       uc->dir_flags = get_be32(&ouc->dir_flags);
+       uc->exclude_per_dir = xstrdup(ouc->exclude_per_dir);
+       /* NUL after exclude_per_dir is covered by sizeof(*ouc) */
+       next += ouc_size(strlen(ouc->exclude_per_dir));
+       if (next >= end)
+               goto done2;
+
+       len = decode_varint(&next);
+       if (next > end || len == 0)
+               goto done2;
+
+       rd.valid      = ewah_new();
+       rd.check_only = ewah_new();
+       rd.sha1_valid = ewah_new();
+       rd.data       = next;
+       rd.end        = end;
+       rd.index      = 0;
+       rd.ucd        = xmalloc(sizeof(*rd.ucd) * len);
+
+       if (read_one_dir(&uc->root, &rd) || rd.index != len)
+               goto done;
+
+       next = rd.data;
+       len = ewah_read_mmap(rd.valid, next, end - next);
+       if (len < 0)
+               goto done;
+
+       next += len;
+       len = ewah_read_mmap(rd.check_only, next, end - next);
+       if (len < 0)
+               goto done;
+
+       next += len;
+       len = ewah_read_mmap(rd.sha1_valid, next, end - next);
+       if (len < 0)
+               goto done;
+
+       ewah_each_bit(rd.check_only, set_check_only, &rd);
+       rd.data = next + len;
+       ewah_each_bit(rd.valid, read_stat, &rd);
+       ewah_each_bit(rd.sha1_valid, read_sha1, &rd);
+       next = rd.data;
+
+done:
+       free(rd.ucd);
+       ewah_free(rd.valid);
+       ewah_free(rd.check_only);
+       ewah_free(rd.sha1_valid);
+done2:
+       if (next != end) {
+               free_untracked_cache(uc);
+               uc = NULL;
+       }
+       return uc;
+}
+
+void untracked_cache_invalidate_path(struct index_state *istate,
+                                    const char *path)
+{
+       const char *sep;
+       struct untracked_cache_dir *d;
+       if (!istate->untracked || !istate->untracked->root)
+               return;
+       sep = strrchr(path, '/');
+       if (sep)
+               d = lookup_untracked(istate->untracked,
+                                    istate->untracked->root,
+                                    path, sep - path);
+       else
+               d = istate->untracked->root;
+       istate->untracked->dir_invalidated++;
+       d->valid = 0;
+       d->untracked_nr = 0;
+}
+
+void untracked_cache_remove_from_index(struct index_state *istate,
+                                      const char *path)
+{
+       untracked_cache_invalidate_path(istate, path);
+}
+
+void untracked_cache_add_to_index(struct index_state *istate,
+                                 const char *path)
+{
+       untracked_cache_invalidate_path(istate, path);
+}
diff --git a/dir.h b/dir.h
index 72b73c65dcfc6f31004a032d0706626424ac2241..7b5855dd80eda02973a55c827d79826a25937880 100644 (file)
--- a/dir.h
+++ b/dir.h
@@ -66,6 +66,7 @@ struct exclude_stack {
        struct exclude_stack *prev; /* the struct exclude_stack for the parent directory */
        int baselen;
        int exclude_ix; /* index of exclude_list within EXC_DIRS exclude_list_group */
+       struct untracked_cache_dir *ucd;
 };
 
 struct exclude_list_group {
@@ -73,6 +74,73 @@ struct exclude_list_group {
        struct exclude_list *el;
 };
 
+struct sha1_stat {
+       struct stat_data stat;
+       unsigned char sha1[20];
+       int valid;
+};
+
+/*
+ *  Untracked cache
+ *
+ *  The following inputs are sufficient to determine what files in a
+ *  directory are excluded:
+ *
+ *   - The list of files and directories of the directory in question
+ *   - The $GIT_DIR/index
+ *   - dir_struct flags
+ *   - The content of $GIT_DIR/info/exclude
+ *   - The content of core.excludesfile
+ *   - The content (or the lack) of .gitignore of all parent directories
+ *     from $GIT_WORK_TREE
+ *   - The check_only flag in read_directory_recursive (for
+ *     DIR_HIDE_EMPTY_DIRECTORIES)
+ *
+ *  The first input can be checked using directory mtime. In many
+ *  filesystems, directory mtime (stat_data field) is updated when its
+ *  files or direct subdirs are added or removed.
+ *
+ *  The second one can be hooked from cache_tree_invalidate_path().
+ *  Whenever a file (or a submodule) is added or removed from a
+ *  directory, we invalidate that directory.
+ *
+ *  The remaining inputs are easy, their SHA-1 could be used to verify
+ *  their contents (exclude_sha1[], info_exclude_sha1[] and
+ *  excludes_file_sha1[])
+ */
+struct untracked_cache_dir {
+       struct untracked_cache_dir **dirs;
+       char **untracked;
+       struct stat_data stat_data;
+       unsigned int untracked_alloc, dirs_nr, dirs_alloc;
+       unsigned int untracked_nr;
+       unsigned int check_only : 1;
+       /* all data except 'dirs' in this struct are good */
+       unsigned int valid : 1;
+       unsigned int recurse : 1;
+       /* null SHA-1 means this directory does not have .gitignore */
+       unsigned char exclude_sha1[20];
+       char name[FLEX_ARRAY];
+};
+
+struct untracked_cache {
+       struct sha1_stat ss_info_exclude;
+       struct sha1_stat ss_excludes_file;
+       const char *exclude_per_dir;
+       struct strbuf ident;
+       /*
+        * dir_struct#flags must match dir_flags or the untracked
+        * cache is ignored.
+        */
+       unsigned dir_flags;
+       struct untracked_cache_dir *root;
+       /* Statistics */
+       int dir_created;
+       int gitignore_invalidated;
+       int dir_invalidated;
+       int dir_opened;
+};
+
 struct dir_struct {
        int nr, alloc;
        int ignored_nr, ignored_alloc;
@@ -120,6 +188,12 @@ struct dir_struct {
        struct exclude_stack *exclude_stack;
        struct exclude *exclude;
        struct strbuf basebuf;
+
+       /* Enable untracked file cache if set */
+       struct untracked_cache *untracked;
+       struct sha1_stat ss_info_exclude;
+       struct sha1_stat ss_excludes_file;
+       unsigned unmanaged_exclude_files;
 };
 
 /*
@@ -226,4 +300,12 @@ static inline int dir_path_match(const struct dir_entry *ent,
                              has_trailing_dir);
 }
 
+void untracked_cache_invalidate_path(struct index_state *, const char *);
+void untracked_cache_remove_from_index(struct index_state *, const char *);
+void untracked_cache_add_to_index(struct index_state *, const char *);
+
+void free_untracked_cache(struct untracked_cache *);
+struct untracked_cache *read_untracked_extension(const void *data, unsigned long sz);
+void write_untracked_extension(struct strbuf *out, struct untracked_cache *untracked);
+void add_untracked_ident(struct untracked_cache *);
 #endif
index 1c2d7afd4cb9b70a324d355c3d69b732b181012b..43481b9c60c8afe30d6916650dd0e765065715c9 100644 (file)
@@ -19,6 +19,7 @@
  */
 #include "git-compat-util.h"
 #include "ewok.h"
+#include "strbuf.h"
 
 int ewah_serialize_native(struct ewah_bitmap *self, int fd)
 {
@@ -110,6 +111,18 @@ int ewah_serialize(struct ewah_bitmap *self, int fd)
        return ewah_serialize_to(self, write_helper, (void *)(intptr_t)fd);
 }
 
+static int write_strbuf(void *user_data, const void *data, size_t len)
+{
+       struct strbuf *sb = user_data;
+       strbuf_add(sb, data, len);
+       return len;
+}
+
+int ewah_serialize_strbuf(struct ewah_bitmap *self, struct strbuf *sb)
+{
+       return ewah_serialize_to(self, write_strbuf, sb);
+}
+
 int ewah_read_mmap(struct ewah_bitmap *self, const void *map, size_t len)
 {
        const uint8_t *ptr = map;
index 13c6e20412591ed3bc56b38b17419a540264f6c5..e73252536702aaf9fed17757937fbaf4b4593f91 100644 (file)
@@ -30,6 +30,7 @@
 #      define ewah_calloc xcalloc
 #endif
 
+struct strbuf;
 typedef uint64_t eword_t;
 #define BITS_IN_WORD (sizeof(eword_t) * 8)
 
@@ -98,6 +99,7 @@ int ewah_serialize_to(struct ewah_bitmap *self,
                      void *out);
 int ewah_serialize(struct ewah_bitmap *self, int fd);
 int ewah_serialize_native(struct ewah_bitmap *self, int fd);
+int ewah_serialize_strbuf(struct ewah_bitmap *self, struct strbuf *);
 
 int ewah_deserialize(struct ewah_bitmap *self, int fd);
 int ewah_read_mmap(struct ewah_bitmap *self, const void *map, size_t len);
index b7a97fbe8300532505b9d63dace2a3f76d8e04e2..17584adbd093a0848debcc23b2a73dcf520a8fb4 100644 (file)
 #elif defined(_MSC_VER)
 #include "compat/msvc.h"
 #else
+#include <sys/utsname.h>
 #include <sys/wait.h>
 #include <sys/resource.h>
 #include <sys/socket.h>
index 36ff89f29e5f56a5b3dcfd803f12cd139295b8b8..723d48dddfe58f50b30105689dfeb9bcb388c8f2 100644 (file)
@@ -39,11 +39,12 @@ static struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
 #define CACHE_EXT_TREE 0x54524545      /* "TREE" */
 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
 #define CACHE_EXT_LINK 0x6c696e6b        /* "link" */
+#define CACHE_EXT_UNTRACKED 0x554E5452   /* "UNTR" */
 
 /* changes that can be kept in $GIT_DIR/index (basically all extensions) */
 #define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
                 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
-                SPLIT_INDEX_ORDERED)
+                SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED)
 
 struct index_state the_index;
 static const char *alternate_index_output;
@@ -79,6 +80,7 @@ void rename_index_entry_at(struct index_state *istate, int nr, const char *new_n
        memcpy(new->name, new_name, namelen + 1);
 
        cache_tree_invalidate_path(istate, old->name);
+       untracked_cache_remove_from_index(istate, old->name);
        remove_index_entry_at(istate, nr);
        add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
 }
@@ -270,20 +272,34 @@ static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
        return changed;
 }
 
-static int is_racy_timestamp(const struct index_state *istate,
-                            const struct cache_entry *ce)
+static int is_racy_stat(const struct index_state *istate,
+                       const struct stat_data *sd)
 {
-       return (!S_ISGITLINK(ce->ce_mode) &&
-               istate->timestamp.sec &&
+       return (istate->timestamp.sec &&
 #ifdef USE_NSEC
                 /* nanosecond timestamped files can also be racy! */
-               (istate->timestamp.sec < ce->ce_stat_data.sd_mtime.sec ||
-                (istate->timestamp.sec == ce->ce_stat_data.sd_mtime.sec &&
-                 istate->timestamp.nsec <= ce->ce_stat_data.sd_mtime.nsec))
+               (istate->timestamp.sec < sd->sd_mtime.sec ||
+                (istate->timestamp.sec == sd->sd_mtime.sec &&
+                 istate->timestamp.nsec <= sd->sd_mtime.nsec))
 #else
-               istate->timestamp.sec <= ce->ce_stat_data.sd_mtime.sec
+               istate->timestamp.sec <= sd->sd_mtime.sec
 #endif
-                );
+               );
+}
+
+static int is_racy_timestamp(const struct index_state *istate,
+                            const struct cache_entry *ce)
+{
+       return (!S_ISGITLINK(ce->ce_mode) &&
+               is_racy_stat(istate, &ce->ce_stat_data));
+}
+
+int match_stat_data_racy(const struct index_state *istate,
+                        const struct stat_data *sd, struct stat *st)
+{
+       if (is_racy_stat(istate, sd))
+               return MTIME_CHANGED;
+       return match_stat_data(sd, st);
 }
 
 int ie_match_stat(const struct index_state *istate,
@@ -538,6 +554,7 @@ int remove_file_from_index(struct index_state *istate, const char *path)
        if (pos < 0)
                pos = -pos-1;
        cache_tree_invalidate_path(istate, path);
+       untracked_cache_remove_from_index(istate, path);
        while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
                remove_index_entry_at(istate, pos);
        return 0;
@@ -982,6 +999,8 @@ static int add_index_entry_with_check(struct index_state *istate, struct cache_e
        }
        pos = -pos-1;
 
+       untracked_cache_add_to_index(istate, ce->name);
+
        /*
         * Inserting a merged entry ("stage 0") into the index
         * will always replace all non-merged entries..
@@ -1372,6 +1391,9 @@ static int read_index_extension(struct index_state *istate,
                if (read_link_extension(istate, data, sz))
                        return -1;
                break;
+       case CACHE_EXT_UNTRACKED:
+               istate->untracked = read_untracked_extension(data, sz);
+               break;
        default:
                if (*ext < 'A' || 'Z' < *ext)
                        return error("index uses %.4s extension, which we do not understand",
@@ -1667,6 +1689,8 @@ int discard_index(struct index_state *istate)
        istate->cache = NULL;
        istate->cache_alloc = 0;
        discard_split_index(istate);
+       free_untracked_cache(istate->untracked);
+       istate->untracked = NULL;
        return 0;
 }
 
@@ -2053,6 +2077,17 @@ static int do_write_index(struct index_state *istate, int newfd,
                if (err)
                        return -1;
        }
+       if (!strip_extensions && istate->untracked) {
+               struct strbuf sb = STRBUF_INIT;
+
+               write_untracked_extension(&sb, istate->untracked);
+               err = write_index_ext_header(&c, newfd, CACHE_EXT_UNTRACKED,
+                                            sb.len) < 0 ||
+                       ce_write(&c, newfd, sb.buf, sb.len) < 0;
+               strbuf_release(&sb);
+               if (err)
+                       return -1;
+       }
 
        if (ce_flush(&c, newfd, istate->sha1) || fstat(newfd, &st))
                return -1;
index 21485e20665979458fb794ee4c3cc77c27032fa8..968b780a06d1f17b190395f06f78fc3124fcf445 100644 (file)
@@ -41,13 +41,6 @@ int read_link_extension(struct index_state *istate,
        return 0;
 }
 
-static int write_strbuf(void *user_data, const void *data, size_t len)
-{
-       struct strbuf *sb = user_data;
-       strbuf_add(sb, data, len);
-       return len;
-}
-
 int write_link_extension(struct strbuf *sb,
                         struct index_state *istate)
 {
@@ -55,8 +48,8 @@ int write_link_extension(struct strbuf *sb,
        strbuf_add(sb, si->base_sha1, 20);
        if (!si->delete_bitmap && !si->replace_bitmap)
                return 0;
-       ewah_serialize_to(si->delete_bitmap, write_strbuf, sb);
-       ewah_serialize_to(si->replace_bitmap, write_strbuf, sb);
+       ewah_serialize_strbuf(si->delete_bitmap, sb);
+       ewah_serialize_strbuf(si->replace_bitmap, sb);
        return 0;
 }
 
diff --git a/t/t7063-status-untracked-cache.sh b/t/t7063-status-untracked-cache.sh
new file mode 100755 (executable)
index 0000000..2b2ffd7
--- /dev/null
@@ -0,0 +1,353 @@
+#!/bin/sh
+
+test_description='test untracked cache'
+
+. ./test-lib.sh
+
+avoid_racy() {
+       sleep 1
+}
+
+git update-index --untracked-cache
+# It's fine if git update-index returns an error code other than one,
+# it'll be caught in the first test.
+if test $? -eq 1; then
+       skip_all='This system does not support untracked cache'
+       test_done
+fi
+
+test_expect_success 'setup' '
+       git init worktree &&
+       cd worktree &&
+       mkdir done dtwo dthree &&
+       touch one two three done/one dtwo/two dthree/three &&
+       git add one two done/one &&
+       : >.git/info/exclude &&
+       git update-index --untracked-cache
+'
+
+test_expect_success 'untracked cache is empty' '
+       test-dump-untracked-cache >../actual &&
+       cat >../expect <<EOF &&
+info/exclude 0000000000000000000000000000000000000000
+core.excludesfile 0000000000000000000000000000000000000000
+exclude_per_dir .gitignore
+flags 00000006
+EOF
+       test_cmp ../expect ../actual
+'
+
+cat >../status.expect <<EOF &&
+A  done/one
+A  one
+A  two
+?? dthree/
+?? dtwo/
+?? three
+EOF
+
+cat >../dump.expect <<EOF &&
+info/exclude e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
+core.excludesfile 0000000000000000000000000000000000000000
+exclude_per_dir .gitignore
+flags 00000006
+/ 0000000000000000000000000000000000000000 recurse valid
+dthree/
+dtwo/
+three
+/done/ 0000000000000000000000000000000000000000 recurse valid
+/dthree/ 0000000000000000000000000000000000000000 recurse check_only valid
+three
+/dtwo/ 0000000000000000000000000000000000000000 recurse check_only valid
+two
+EOF
+
+test_expect_success 'status first time (empty cache)' '
+       avoid_racy &&
+       : >../trace &&
+       GIT_TRACE_UNTRACKED_STATS="$TRASH_DIRECTORY/trace" \
+       git status --porcelain >../actual &&
+       test_cmp ../status.expect ../actual &&
+       cat >../trace.expect <<EOF &&
+node creation: 3
+gitignore invalidation: 1
+directory invalidation: 0
+opendir: 4
+EOF
+       test_cmp ../trace.expect ../trace
+'
+
+test_expect_success 'untracked cache after first status' '
+       test-dump-untracked-cache >../actual &&
+       test_cmp ../dump.expect ../actual
+'
+
+test_expect_success 'status second time (fully populated cache)' '
+       avoid_racy &&
+       : >../trace &&
+       GIT_TRACE_UNTRACKED_STATS="$TRASH_DIRECTORY/trace" \
+       git status --porcelain >../actual &&
+       test_cmp ../status.expect ../actual &&
+       cat >../trace.expect <<EOF &&
+node creation: 0
+gitignore invalidation: 0
+directory invalidation: 0
+opendir: 0
+EOF
+       test_cmp ../trace.expect ../trace
+'
+
+test_expect_success 'untracked cache after second status' '
+       test-dump-untracked-cache >../actual &&
+       test_cmp ../dump.expect ../actual
+'
+
+test_expect_success 'modify in root directory, one dir invalidation' '
+       avoid_racy &&
+       : >four &&
+       : >../trace &&
+       GIT_TRACE_UNTRACKED_STATS="$TRASH_DIRECTORY/trace" \
+       git status --porcelain >../actual &&
+       cat >../status.expect <<EOF &&
+A  done/one
+A  one
+A  two
+?? dthree/
+?? dtwo/
+?? four
+?? three
+EOF
+       test_cmp ../status.expect ../actual &&
+       cat >../trace.expect <<EOF &&
+node creation: 0
+gitignore invalidation: 0
+directory invalidation: 1
+opendir: 1
+EOF
+       test_cmp ../trace.expect ../trace
+
+'
+
+test_expect_success 'verify untracked cache dump' '
+       test-dump-untracked-cache >../actual &&
+       cat >../expect <<EOF &&
+info/exclude e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
+core.excludesfile 0000000000000000000000000000000000000000
+exclude_per_dir .gitignore
+flags 00000006
+/ 0000000000000000000000000000000000000000 recurse valid
+dthree/
+dtwo/
+four
+three
+/done/ 0000000000000000000000000000000000000000 recurse valid
+/dthree/ 0000000000000000000000000000000000000000 recurse check_only valid
+three
+/dtwo/ 0000000000000000000000000000000000000000 recurse check_only valid
+two
+EOF
+       test_cmp ../expect ../actual
+'
+
+test_expect_success 'new .gitignore invalidates recursively' '
+       avoid_racy &&
+       echo four >.gitignore &&
+       : >../trace &&
+       GIT_TRACE_UNTRACKED_STATS="$TRASH_DIRECTORY/trace" \
+       git status --porcelain >../actual &&
+       cat >../status.expect <<EOF &&
+A  done/one
+A  one
+A  two
+?? .gitignore
+?? dthree/
+?? dtwo/
+?? three
+EOF
+       test_cmp ../status.expect ../actual &&
+       cat >../trace.expect <<EOF &&
+node creation: 0
+gitignore invalidation: 1
+directory invalidation: 1
+opendir: 4
+EOF
+       test_cmp ../trace.expect ../trace
+
+'
+
+test_expect_success 'verify untracked cache dump' '
+       test-dump-untracked-cache >../actual &&
+       cat >../expect <<EOF &&
+info/exclude e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
+core.excludesfile 0000000000000000000000000000000000000000
+exclude_per_dir .gitignore
+flags 00000006
+/ e6fcc8f2ee31bae321d66afd183fcb7237afae6e recurse valid
+.gitignore
+dthree/
+dtwo/
+three
+/done/ 0000000000000000000000000000000000000000 recurse valid
+/dthree/ 0000000000000000000000000000000000000000 recurse check_only valid
+three
+/dtwo/ 0000000000000000000000000000000000000000 recurse check_only valid
+two
+EOF
+       test_cmp ../expect ../actual
+'
+
+test_expect_success 'new info/exclude invalidates everything' '
+       avoid_racy &&
+       echo three >>.git/info/exclude &&
+       : >../trace &&
+       GIT_TRACE_UNTRACKED_STATS="$TRASH_DIRECTORY/trace" \
+       git status --porcelain >../actual &&
+       cat >../status.expect <<EOF &&
+A  done/one
+A  one
+A  two
+?? .gitignore
+?? dtwo/
+EOF
+       test_cmp ../status.expect ../actual &&
+       cat >../trace.expect <<EOF &&
+node creation: 0
+gitignore invalidation: 1
+directory invalidation: 0
+opendir: 4
+EOF
+       test_cmp ../trace.expect ../trace
+'
+
+test_expect_success 'verify untracked cache dump' '
+       test-dump-untracked-cache >../actual &&
+       cat >../expect <<EOF &&
+info/exclude 13263c0978fb9fad16b2d580fb800b6d811c3ff0
+core.excludesfile 0000000000000000000000000000000000000000
+exclude_per_dir .gitignore
+flags 00000006
+/ e6fcc8f2ee31bae321d66afd183fcb7237afae6e recurse valid
+.gitignore
+dtwo/
+/done/ 0000000000000000000000000000000000000000 recurse valid
+/dthree/ 0000000000000000000000000000000000000000 recurse check_only valid
+/dtwo/ 0000000000000000000000000000000000000000 recurse check_only valid
+two
+EOF
+       test_cmp ../expect ../actual
+'
+
+test_expect_success 'move two from tracked to untracked' '
+       git rm --cached two &&
+       test-dump-untracked-cache >../actual &&
+       cat >../expect <<EOF &&
+info/exclude 13263c0978fb9fad16b2d580fb800b6d811c3ff0
+core.excludesfile 0000000000000000000000000000000000000000
+exclude_per_dir .gitignore
+flags 00000006
+/ e6fcc8f2ee31bae321d66afd183fcb7237afae6e recurse
+/done/ 0000000000000000000000000000000000000000 recurse valid
+/dthree/ 0000000000000000000000000000000000000000 recurse check_only valid
+/dtwo/ 0000000000000000000000000000000000000000 recurse check_only valid
+two
+EOF
+       test_cmp ../expect ../actual
+'
+
+test_expect_success 'status after the move' '
+       : >../trace &&
+       GIT_TRACE_UNTRACKED_STATS="$TRASH_DIRECTORY/trace" \
+       git status --porcelain >../actual &&
+       cat >../status.expect <<EOF &&
+A  done/one
+A  one
+?? .gitignore
+?? dtwo/
+?? two
+EOF
+       test_cmp ../status.expect ../actual &&
+       cat >../trace.expect <<EOF &&
+node creation: 0
+gitignore invalidation: 0
+directory invalidation: 0
+opendir: 1
+EOF
+       test_cmp ../trace.expect ../trace
+'
+
+test_expect_success 'verify untracked cache dump' '
+       test-dump-untracked-cache >../actual &&
+       cat >../expect <<EOF &&
+info/exclude 13263c0978fb9fad16b2d580fb800b6d811c3ff0
+core.excludesfile 0000000000000000000000000000000000000000
+exclude_per_dir .gitignore
+flags 00000006
+/ e6fcc8f2ee31bae321d66afd183fcb7237afae6e recurse valid
+.gitignore
+dtwo/
+two
+/done/ 0000000000000000000000000000000000000000 recurse valid
+/dthree/ 0000000000000000000000000000000000000000 recurse check_only valid
+/dtwo/ 0000000000000000000000000000000000000000 recurse check_only valid
+two
+EOF
+       test_cmp ../expect ../actual
+'
+
+test_expect_success 'move two from untracked to tracked' '
+       git add two &&
+       test-dump-untracked-cache >../actual &&
+       cat >../expect <<EOF &&
+info/exclude 13263c0978fb9fad16b2d580fb800b6d811c3ff0
+core.excludesfile 0000000000000000000000000000000000000000
+exclude_per_dir .gitignore
+flags 00000006
+/ e6fcc8f2ee31bae321d66afd183fcb7237afae6e recurse
+/done/ 0000000000000000000000000000000000000000 recurse valid
+/dthree/ 0000000000000000000000000000000000000000 recurse check_only valid
+/dtwo/ 0000000000000000000000000000000000000000 recurse check_only valid
+two
+EOF
+       test_cmp ../expect ../actual
+'
+
+test_expect_success 'status after the move' '
+       : >../trace &&
+       GIT_TRACE_UNTRACKED_STATS="$TRASH_DIRECTORY/trace" \
+       git status --porcelain >../actual &&
+       cat >../status.expect <<EOF &&
+A  done/one
+A  one
+A  two
+?? .gitignore
+?? dtwo/
+EOF
+       test_cmp ../status.expect ../actual &&
+       cat >../trace.expect <<EOF &&
+node creation: 0
+gitignore invalidation: 0
+directory invalidation: 0
+opendir: 1
+EOF
+       test_cmp ../trace.expect ../trace
+'
+
+test_expect_success 'verify untracked cache dump' '
+       test-dump-untracked-cache >../actual &&
+       cat >../expect <<EOF &&
+info/exclude 13263c0978fb9fad16b2d580fb800b6d811c3ff0
+core.excludesfile 0000000000000000000000000000000000000000
+exclude_per_dir .gitignore
+flags 00000006
+/ e6fcc8f2ee31bae321d66afd183fcb7237afae6e recurse valid
+.gitignore
+dtwo/
+/done/ 0000000000000000000000000000000000000000 recurse valid
+/dthree/ 0000000000000000000000000000000000000000 recurse check_only valid
+/dtwo/ 0000000000000000000000000000000000000000 recurse check_only valid
+two
+EOF
+       test_cmp ../expect ../actual
+'
+
+test_done
diff --git a/test-dump-untracked-cache.c b/test-dump-untracked-cache.c
new file mode 100644 (file)
index 0000000..25d855d
--- /dev/null
@@ -0,0 +1,62 @@
+#include "cache.h"
+#include "dir.h"
+
+static int compare_untracked(const void *a_, const void *b_)
+{
+       const char *const *a = a_;
+       const char *const *b = b_;
+       return strcmp(*a, *b);
+}
+
+static int compare_dir(const void *a_, const void *b_)
+{
+       const struct untracked_cache_dir *const *a = a_;
+       const struct untracked_cache_dir *const *b = b_;
+       return strcmp((*a)->name, (*b)->name);
+}
+
+static void dump(struct untracked_cache_dir *ucd, struct strbuf *base)
+{
+       int i, len;
+       qsort(ucd->untracked, ucd->untracked_nr, sizeof(*ucd->untracked),
+             compare_untracked);
+       qsort(ucd->dirs, ucd->dirs_nr, sizeof(*ucd->dirs),
+             compare_dir);
+       len = base->len;
+       strbuf_addf(base, "%s/", ucd->name);
+       printf("%s %s", base->buf,
+              sha1_to_hex(ucd->exclude_sha1));
+       if (ucd->recurse)
+               fputs(" recurse", stdout);
+       if (ucd->check_only)
+               fputs(" check_only", stdout);
+       if (ucd->valid)
+               fputs(" valid", stdout);
+       printf("\n");
+       for (i = 0; i < ucd->untracked_nr; i++)
+               printf("%s\n", ucd->untracked[i]);
+       for (i = 0; i < ucd->dirs_nr; i++)
+               dump(ucd->dirs[i], base);
+       strbuf_setlen(base, len);
+}
+
+int main(int ac, char **av)
+{
+       struct untracked_cache *uc;
+       struct strbuf base = STRBUF_INIT;
+       setup_git_directory();
+       if (read_cache() < 0)
+               die("unable to read index file");
+       uc = the_index.untracked;
+       if (!uc) {
+               printf("no untracked cache\n");
+               return 0;
+       }
+       printf("info/exclude %s\n", sha1_to_hex(uc->ss_info_exclude.sha1));
+       printf("core.excludesfile %s\n", sha1_to_hex(uc->ss_excludes_file.sha1));
+       printf("exclude_per_dir %s\n", uc->exclude_per_dir);
+       printf("flags %08x\n", uc->dir_flags);
+       if (uc->root)
+               dump(uc->root, &base);
+       return 0;
+}
index be84ba2607ad2dbdb17397869c459418abad78e4..2927660d929eee776d43a87851a928df12b17716 100644 (file)
@@ -9,6 +9,7 @@
 #include "refs.h"
 #include "attr.h"
 #include "split-index.h"
+#include "dir.h"
 
 /*
  * Error messages expected by scripts out of plumbing commands such as
@@ -1259,8 +1260,10 @@ static int verify_uptodate_sparse(const struct cache_entry *ce,
 static void invalidate_ce_path(const struct cache_entry *ce,
                               struct unpack_trees_options *o)
 {
-       if (ce)
-               cache_tree_invalidate_path(o->src_index, ce->name);
+       if (!ce)
+               return;
+       cache_tree_invalidate_path(o->src_index, ce->name);
+       untracked_cache_invalidate_path(o->src_index, ce->name);
 }
 
 /*
index 38cb165f124d610c789a8d82de42281c795111f7..33452f169dfca7a3cd4f15b1c3cdc91a3cf40735 100644 (file)
@@ -585,6 +585,8 @@ static void wt_status_collect_untracked(struct wt_status *s)
                        DIR_SHOW_OTHER_DIRECTORIES | DIR_HIDE_EMPTY_DIRECTORIES;
        if (s->show_ignored_files)
                dir.flags |= DIR_SHOW_IGNORED_TOO;
+       else
+               dir.untracked = the_index.untracked;
        setup_standard_excludes(&dir);
 
        fill_directory(&dir, &s->pathspec);